Questions about Fritz Benchmark

[Daniel Charles]

does someone know how the Fritz benchmark work?

this is mainly based on the CPU

question:

1/will it change a lot if we add more memory ??
2/will it change with a bigger graphic card ?


From Jens Benchmark two computer with the same cpu don t have the same Kn/s ...

Daniel

[Eelco de Groot]

does someone know how the Fritz benchmark work?

this is mainly based on the CPU

question:

1/will it change a lot if we add more memory ??
2/will it change with a bigger graphic card ?


From Jens Benchmark two computer with the same cpu don t have the same Kn/s ...

Daniel

Hello Daniel,

Maybe the Fritz experts can tell you more about this, I think the Fritz benchmark can basically only say something about the potential speed-up of the program in nodes per second based on number of processors and clock-rate of the CPUs. It does not take memory speeds or effects of graphic cards into account I think, apart probably from the FSB (Front Side Bus) speed? I don't think it can say much about the effective speed-up of a program running on a multi-CPU compared to a single CPU, issues like scaling, the loss of efficiency that takes place because some positions end up calculated twice or more etc. in a multithreaded or SMP program, are not taken into account.

Regards, Eelco

[M ANSARI]

I find it gives a general good indication of the power of a system in playing chess. I noticed that it particularly likes low latencies and high front side bus ... which turns out to also to make Rybka play stronger. If you want to tune your score for Fritzmark ... you have to try to get the fastest FSB and lowest latencies at a given CPU speed. So a 10 x 400 = 4Ghz will score better than a 11 x 364 = 4Ghz

[geots]

does someone know how the Fritz benchmark work?

this is mainly based on the CPU

question:

1/will it change a lot if we add more memory ??
2/will it change with a bigger graphic card ?


From Jens Benchmark two computer with the same cpu don t have the same Kn/s ...

Daniel

Hi, Daniel. I got this from the experts at chessbase, whatever that means. But i did some more checking with the guy who invented the benchmark. He told me that its ONE AND ONLY use is to compare certain things between 2 computers. Im no expert, and i forget what he said you derived from the benchmark. But he said it has no use and is no good for things like determining the best hash to use for an engine. He said he tried to convince people of this, but even Steve Lopez who writes articles for chessbase uses it to help determine hash. He said it is worthless for that, but people would not listen to him and kept trying to use it for things that it wasnt designed for and got faulty assumptions. The following is not my words, but a direct quote from the benchmark inventor himself- "Try as i might, i could not convince people it was useless to use on just on one computer, and told you nothing. It is only good for comparing certain things between 2 computers. But try as i might- they would not listen. So i finally said the hell with it- they were convinced it helped with things like determing hash. I gave up trying to convince the son of a bitches. Let them think what they want. I dont care any longer." I dont know if this helped you any- probably not. But it is true and interesting.

Regards,

George

[AGove]

Here is a page with a table of the benchmarks: http://www.jens.tauchclub-krems.at/dive ... marks.html

I use the benchmark solely to comfort myself that all the money I've spent on a new PC was justified, that my friends' PCs are vastly inferior, and so on. But nothing scientific.

[ernest]

I use the Fritz Benchmark as a standalone application.
On my system it uses 266500 KB memory.
So it obviously works with a fixed amount of hash (over which I have no control) and the graphic card is irrelevant.

[Spock]

I find it gives a general good indication of the power of a system in playing chess. I noticed that it particularly likes low latencies and high front side bus ... which turns out to also to make Rybka play stronger. If you want to tune your score for Fritzmark ... you have to try to get the fastest FSB and lowest latencies at a given CPU speed. So a 10 x 400 = 4Ghz will score better than a 11 x 364 = 4Ghz

Actually I noticed the exact opposite - it doesn't seem to chaneg much if at all with different FSB and timings. Worse still, the results are not repeatable between runs, so it's useless as a benchmark. I did a few re-boots and ran it straight away after each, and the results vary. So I concluded it was useless for system tuning